Dynamic Soft Sensor Modeling Method Fusing Process Feature Information Based on an Improved Intelligent Optimization Algorithm
Bailun Zhang,Jing Zhang,Yongming Han,Zhiqiang Geng
DOI: https://doi.org/10.1016/j.chemolab.2021.104415
IF: 4.175
2021-01-01
Chemometrics and Intelligent Laboratory Systems
Abstract:The lack of real-time and accurate melting index forecast model puts forward challenges to the online quality detection and control of the polypropylene industrial production. Meanwhile, the key process characteristics of dynamic and time delay are not fully and effectively integrated into the soft sensor model. Therefore, a novel dynamic soft sensing (DSS) modeling method using the orthogonal echo state network (OESN) based on the cumulative mutual information (CMI) threshold setting integrating an improved differential evolution algorithm (IDE) (OESN-CMI-IDE) is proposed. The mutual information of each candidate auxiliary variable relative to the dominant variable is calculated and sorted to determine the CMI threshold and select the corresponding auxiliary variable. Then an IDE algorithm is used to optimize the parameters of the OESN model, which has faster convergence, higher accuracy and stronger robustness than that of DE, self-adaptive DE (jDE), DE with time varying scale factor (DETVSF), line decreased inertia weight particle swarm optimization (LDPSO), flexible exponential inertia weight particle swarm optimization (FEPSO), and oscillating triangular inertia weight particle swarm optimization (OTPSO) in terms of evolutionary computing competition benchmark function tests. Finally, the proposed method is applied to predict the melt index in an actual polypropylene plant. The experimental results demonstrate that the prediction effect of the proposed method is better than that of the OESN model optimized by DE, jDE, DETVSF, LDPSO, FEPSO, OTPSO and the OESN model without optimization, which can escort the online monitoring and advanced control of the melt index in the propylene industry.